Examining Life Amidst the COVID Era

summary(df)

##                Region         Age        OnlineClass_Time OnlineClass_Rating
##  Delhi-NCR        :721   Min.   : 7.00   Min.   : 0.000   Very poor:437     
##  Outside Delhi-NCR:461   1st Qu.:17.00   1st Qu.: 2.000   Poor     : 30     
##                          Median :20.00   Median : 3.000   Average  :387     
##                          Mean   :20.17   Mean   : 3.209   Good     :230     
##                          3rd Qu.:21.00   3rd Qu.: 5.000   Excellent: 98     
##                          Max.   :59.00   Max.   :10.000                     
##                                                                             
##                     OnlineClass_Medium SelfStudy_Duration Fitness_Duration
##  Laptop/Desktop              :596      Min.   : 0.000     Min.   :0.0000  
##  Smartphone                  :539      1st Qu.: 2.000     1st Qu.:0.0000  
##  Tablet                      : 37      Median : 2.000     Median :1.0000  
##  Any Gadget                  :  5      Mean   : 2.912     Mean   :0.7658  
##  Smartphone or Laptop/Desktop:  5      3rd Qu.: 4.000     3rd Qu.:1.0000  
##                                        Max.   :18.000     Max.   :5.0000  
##                                                                           
##  Sleeping_Duration SosMed_Duration    SosMed_Medium WatchingTV_Duration
##  Min.   : 4.000    Min.   : 0.000   Instagram:370   Min.   : 0.000     
##  1st Qu.: 7.000    1st Qu.: 1.000   WhatsApp :337   1st Qu.: 0.000     
##  Median : 8.000    Median : 2.000   Youtube  :314   Median : 1.000     
##  Mean   : 7.871    Mean   : 2.366   Linkedin : 61   Mean   : 1.023     
##  3rd Qu.: 9.000    3rd Qu.: 3.000   Facebook : 52   3rd Qu.: 2.000     
##  Max.   :15.000    Max.   :10.000   Twitter  : 28   Max.   :15.000     
##                                     (Other)  : 20                      
##  NumberOfMeals            WeightChange Having_HealthIssue
##  Min.   :1.000   Decreased      :209   NO :1021          
##  1st Qu.:2.000   Remain Constant:535   YES: 161          
##  Median :3.000   Increased      :438                     
##  Mean   :2.918                                           
##  3rd Qu.:3.000                                           
##  Max.   :8.000                                           
##                                                          
##                         Stress_Busters Time_Utilized
##  Listening to music            :276    NO :608      
##  Online gaming                 :175    YES:574      
##  Watching web series           :102                 
##  Reading books                 : 77                 
##  Scrolling through social media: 74                 
##  Sleeping                      : 71                 
##  (Other)                       :407                 
##  Connected_with_Family_Friends_Relatives             Most_Missed_Things
##  NO :351                                 School/college       :379     
##  YES:831                                 Friends , relatives  :223     
##                                          Travelling           :183     
##                                          Roaming around freely:149     
##                                          Eating outside       :104     
##                                          Colleagues           : 67     
##                                          (Other)              : 77

Categorical Variables:

The summary shows how often each category was chosen for variables like Region, OnlineClass_Rating, OnlineClass_Medium, and others.

• Region: The dataset includes 461 participants from places outside Delhi-NCR and 721 participants from Delhi-NCR.
• OnlineClass_Rating: Ratings for online classes varied among participants, with responses ranging from very poor (437), poor (30), average (387), good (230), to excellent (98), reflecting diverse opinions on online learning experiences.
• OnlineClass_Medium: Participants used different devices for online classes, such as laptop/desktop (596), smartphone (539), tablet (37), any gadget (5), and both smartphone and laptop/desktop (5), indicating preferences for flexible learning methods.
• SosMed_Medium: Various social media platforms were used by participants, including Instagram (370), WhatsApp (337), YouTube (314), LinkedIn (61), Facebook (52), Twitter (28), and other platforms (20), highlighting diverse social media usage.
• WeightChange: Participants reported weight changes during the pandemic, with decreases (209), no change (535), and increases (438), showing varying impacts on physical health.
• Having_HealthIssue: Participants reported health issues, with responses split into no (1021) and yes (161), underscoring the prevalence of health problems during the pandemic.
• Stress_Busters: Coping strategies included listening to music (276), playing online video games (175), watching web series (102), reading books (77), browsing social media (74), sleeping (71), and other methods (407), indicating diverse ways of managing stress.
• Time_Utilized: Participants reported whether they utilized their time effectively during the pandemic, with responses categorized as no (608) and yes (574), showing different levels of productivity.
• Connected_with_Family_Friends_Relatives: Participants felt connected with family, friends, or relatives, with responses categorized as no (351) and yes (831), emphasizing the importance of social connections during periods of isolation.
• Most_Missed_Things: Activities missed the most during the pandemic included school/college (379), time with friends/relatives (223), traveling (183), freely roaming (149), dining out (104), colleagues (67), and other activities (77), highlighting the impact of restrictions on daily activities and social interactions.

Numeric Variables:

The summary gives us information about variables such as Age, OnlineClass_Time, SelfStudy_Duration, and more.

• Age: The participants’ ages ranged from 7 to 59 years, with most around 20 years old.
• OnlineClass_Time: During the pandemic, participants spent between 0 to 10 hours daily on online classes, with an average of 3.21 hours.
• SelfStudy_Duration: Study habits varied widely, with participants spending from 0 to 18 hours daily on self-study, averaging 2.91 hours.
• Fitness_Duration: Daily physical activity varied, with participants engaging from 0 to 5 hours, averaging 0.77 hours.
• Sleeping_Duration: Participants slept an average of 7.87 hours daily, with most sleeping around 8 hours.
• SosMed_Duration: Daily social media use ranged from 0 to 10 hours, with an average of 2.37 hours.
• WatchingTV_Duration: Daily TV viewing ranged from 0 to 15 hours, averaging 1.02 hours.
• NumberOfMeals: Participants typically ate 2.92 meals daily, with most having around 3 meals.

These summaries provide insights into how participants spent their time and managed daily activities during the pandemic.

Note: (Other) combine all remaining levels of a factor variable with more than six levels into a single “other” category Numeric variables provide mean value and Tukey’s five-number summary which includes the minimum, first quartile (Q1), median (Q2), third quartile (Q3), and maximum

BasicSummary(df)

##                                   variable    type levels           topLevel
## 1                                   Region  factor      2          Delhi-NCR
## 2                                      Age numeric     42                 20
## 3                         OnlineClass_Time numeric     21                  4
## 4                       OnlineClass_Rating  factor      5          Very poor
## 5                       OnlineClass_Medium  factor      5     Laptop/Desktop
## 6                       SelfStudy_Duration numeric     23                  2
## 7                         Fitness_Duration numeric     11                  1
## 8                        Sleeping_Duration numeric     18                  8
## 9                          SosMed_Duration numeric     22                  1
## 10                           SosMed_Medium  factor     13          Instagram
## 11                     WatchingTV_Duration numeric     22                  0
## 12                           NumberOfMeals numeric      8                  3
## 13                            WeightChange  factor      3    Remain Constant
## 14                      Having_HealthIssue  factor      2                 NO
## 15                          Stress_Busters  factor     86 Listening to music
## 16                           Time_Utilized  factor      2                 NO
## 17 Connected_with_Family_Friends_Relatives  factor      2                YES
## 18                      Most_Missed_Things  factor     36     School/college
##    topCount topFrac missFreq missFrac
## 1       721   0.610        0        0
## 2       211   0.179        0        0
## 3       222   0.188        0        0
## 4       437   0.370        0        0
## 5       596   0.504        0        0
## 6       346   0.293        0        0
## 7       550   0.465        0        0
## 8       390   0.330        0        0
## 9       343   0.290        0        0
## 10      370   0.313        0        0
## 11      444   0.376        0        0
## 12      610   0.516        0        0
## 13      535   0.453        0        0
## 14     1021   0.864        0        0
## 15      276   0.234        0        0
## 16      608   0.514        0        0
## 17      831   0.703        0        0
## 18      379   0.321        0        0

• The majority of the variables in this dataset have clear and meaningful names, which is crucial for understanding the data easily
• Among the 18 variables in the df dataset, 10 are categorical factors, and the rest of the columns contain numeric data
  • variable: name or label of the variable
  • type: data type of the variable (ex: character, numeric)
  • levels: number of unique categories/ values
  • topLevel: most frequent category/value
  • topCount: frequency of the top-level category/value
  • topFrac: proportion of the top-level category/value
  • missFreq: count of missing values
  • missFrac: proportion of missing values

num_df <- subset(df,select = num_cols) # make new data frame with consisting only numeric/ integer data type
convert_matrix(num_df)

##                       Age OnlineClass_Time SelfStudy_Duration Fitness_Duration
## Mean                20.17             3.21               2.91             0.77
## Median              20.00             3.00               2.00             1.00
## Variance            30.43             4.42               4.58             0.52
## Standard Deviation   5.52             2.10               2.14             0.72
## Range               52.00            10.00              18.00             5.00
## Interquartile Range  4.00             3.00               2.00             1.00
##                     Sleeping_Duration SosMed_Duration WatchingTV_Duration
## Mean                             7.87            2.37                1.02
## Median                           8.00            2.00                1.00
## Variance                         2.61            3.12                1.60
## Standard Deviation               1.62            1.77                1.26
## Range                           11.00           10.00               15.00
## Interquartile Range              2.00            2.00                2.00
##                     NumberOfMeals
## Mean                         2.92
## Median                       3.00
## Variance                     0.69
## Standard Deviation           0.83
## Range                        7.00
## Interquartile Range          1.00

These statistical measures offer insights into how data is distributed within a dataset:

• Central Tendency: mean and median help to understand the central value of the data.
• Variability: variance and standard deviation measure the spread of the data points around the mean
• Range: quick insight into the spread between the minimum and maximum values
• Distribution and Spread: interquartile range (IQR) provides insights about the spread of the middle 50% of the data, giving a solid measure of the data’s distribution that is less influenced by extreme values or outliers

# Age of Respondent visualization (group to some categories with intervals of 10)
age_plot <- ggplot(df, aes(x = Age, fill=Having_HealthIssue)) + # x axis represents age, and the color will be differentiate by having health issue
  geom_histogram(color = "black", binwidth = 10) + # our histogram will have black outline and the bars are group per 10 intervaols
  # add label to our graph
  labs(title = "Distribution of Respondents by Age",
       x = "Age",
       y = "Number of Respondents",
       fill = "Having Health Issue?") +
  scale_fill_brewer(palette = "Set2") + # filling colors using color palettes from the RColorBrewer package
  coord_flip() # swapping the x and y axis, it is useful when the data categorical name is too long and it doesn't fit in the x axis

ggplotly(age_plot)

The survey responses indicate that most people who answered are teenagers or in their twenties. This tells us important things about the different issues people of different ages might have. Younger students might find it hard to stay focused and disciplined when they’re learning online. Older students might worry about what will happen with their careers in the future.

# Respondent's Region of Residence visualization
residence_plot <- pie(table(df$Region), # table is used to create a contingency table of the counts of each category
    main = "Distribution of Respondents by Region of Residence",
    col = c("purple", "pink"))

The analysis shows that a large majority, more than half of the respondents, live in the Delhi-National Capital Region (Delhi-NCR).

time_plot <- pie(table(df$Time_Utilized),
    main = "Distribution of Respondents by Time Management",
    col = c("purple", "pink"))

The study reveals that time management has been a significant challenge for students during the pandemic. According to the research, 51.44% of students felt they were not effectively using their time. This finding illustrates the struggle students face in adapting to new routines and managing their time well in an online learning setup.

health_plot <- pie(table(df$Having_HealthIssue),
    main = "Distribution of Respondents by Having Health Issue",
    col = c("purple", "pink"))

Based on the survey, about 13.62% of students mentioned having health issues, showing a notable impact on their physical and mental well-being. This finding mirrors a broader worldwide pattern of increased health concerns due to extended lockdowns and lifestyle adjustments.

# Respondent's Preferred Social Media Platform visualization
sosmed_plot <- df %>% # %>% is a pipe operator to chain multiple operation
  group_by(SosMed_Medium) %>% # group data by SosMed_Medium
  summarise(freqCount = n()) %>% # returns one row for each combination of grouping variables; n() -> counts the number of frequency count in each group
  ggplot(aes(x=reorder(SosMed_Medium, freqCount),  y=freqCount, fill=freqCount)) + # reorder the sosmed_medium factor based on freqCount in ascending order, eventhough it looks like descending because coord_flip is used here
  geom_bar(color="black", stat="identity") + # stat = "identity" -> height of the bars is determined by freqCount
  labs(title = "Distribution of Respondents by Preferred Social Media Platform",
       x = "Social Media Platforms",
       y = "Number of Respondents",
       fill = "Frequency Count") +
  scale_fill_gradient(low = "lightblue", high = "darkblue") + # filling colors using gradient with starting color light blue and end with red
  coord_flip()
ggplotly(sosmed_plot)

The report also examines which social media platforms students preferred most, offering insights into their online habits and how social media helped them stay connected during the pandemic. Instagram and WhatsApp emerged as the top choices, highlighting their significant role in students’ social interactions.

# Top 20 Respondent's Preferred Stress Busters
stress_plot <- df %>%
  group_by(Stress_Busters) %>% 
  summarise(freqCount = n()) %>%
  top_n(10, freqCount) %>%  # filter the top 10 stress busters
  ggplot(aes(x=reorder(Stress_Busters, freqCount),  y=freqCount, fill=freqCount)) +
  geom_bar(color="black", stat="identity") +
  labs(title = "Distribution of Respondents by Preferred Stress Busters (Top 10)",
       x = "Activity",
       y = "Number of Respondents",
       fill = "Frequency Count") +
  scale_fill_gradient(low = "lightblue", high = "darkblue") +
  coord_flip()
ggplotly(stress_plot)

Managing stress is another important concern. Throughout the pandemic, students found solace in activities like listening to music and playing online video games, which ranked among their top five ways to cope with stress. These activities played a crucial role in helping students navigate these challenging times by providing relaxation and stress relief.

# Top 10 Things the Respondent Miss the Most
miss_plot <- df %>%
  group_by(Most_Missed_Things) %>% 
  summarise(freqCount = n()) %>%
  top_n(10, freqCount) %>%
  ggplot(aes(x=reorder(Most_Missed_Things, freqCount),  y=freqCount, fill=freqCount)) +
  geom_bar(color="black", stat="identity") +
  labs(title = "Distribution of Respondents by Things they Miss the Most (Top 10)",
       x = "Object/ Subject",
       y = "Number of Respondents",
       fill = "Frequency Count") +
  scale_fill_gradient(low = "lightblue", high = "darkblue") +
  coord_flip()
ggplotly(miss_plot)

Students expressed longing for aspects of their lives before the pandemic, such as attending school and college and spending time with friends and family. These factors highlight the routines and connections students valued most and illustrate how the pandemic disrupted their social and academic environments. The lack of face-to-face interactions and structured schedules had a significant impact on their daily lives.

# store the column name of variables related to duration
duration_cols <- c('OnlineClass_Time', 'SelfStudy_Duration', 'Fitness_Duration',
                   'Sleeping_Duration', 'SosMed_Duration', 'WatchingTV_Duration')

# calculate mean duration for each activity
mean_durations <- sapply(duration_cols, function(col) mean(df[[col]], na.rm = TRUE)) # a function which take the parameter col to calculate each mean
names(mean_durations) <- duration_cols # labeling the vector

# create a long-format data frame
df_long <- stack(df[, duration_cols]) # stack the duration_cols -> resulting a data frame with 2 columns (the value and the names of stacked columns)
names(df_long) <- c("Duration", "Activity")

# reorder activities by mean of duration
ordered_activities <- names(sort(mean_durations))
df_long$Activity <- factor(df_long$Activity, levels = ordered_activities)

# write.table(df_long, file="df_long.csv", col.names = TRUE, row.names = FALSE, sep=",")

# create the box plot
duration_plot <- ggplot(df_long, aes(y = Activity, x = Duration)) +
  geom_boxplot(fill="gray") +
  labs(title = "Distribution of Time Spent by Activity",
       x = "Duration (Hours(s))",
       y = "Type of Activity")

print(duration_plot)

Looking into how students used their time uncovered some concerning patterns. Exercise was the least prioritized activity, suggesting a lack of physical activity. This lack can lead to lasting negative effects on both physical and mental well-being. It’s crucial to encourage children to incorporate more physical activity into their daily schedules.

# remove outliers in duration column based on the outlier we found in the boxplot
for (i in duration_cols)
{
  outliers <- boxplot.stats(df[[i]])$out # identify outliers
  df <- df[!df[[i]] %in% outliers, ] # remove the whole row if it contains outliers
}

str(df)

## 'data.frame':    1048 obs. of  18 variables:
##  $ Region                                 : Factor w/ 2 levels "Delhi-NCR","Outside Delhi-NCR": 1 1 1 1 1 1 1 2 1 2 ...
##  $ Age                                    : num  21 21 20 20 21 21 19 19 21 20 ...
##  $ OnlineClass_Time                       : num  2 0 7 3 3 0 2 2 3 0 ...
##  $ OnlineClass_Rating                     : Factor w/ 5 levels "Very poor","Poor",..: 4 5 1 1 4 1 1 1 1 1 ...
##  $ OnlineClass_Medium                     : Factor w/ 5 levels "Laptop/Desktop",..: 1 2 1 2 1 2 2 3 1 1 ...
##  $ SelfStudy_Duration                     : num  4 0 3 2 3 6 2 1 4 1 ...
##  $ Fitness_Duration                       : num  0 2 0 1 1 0 1 1 1 0.5 ...
##  $ Sleeping_Duration                      : num  7 10 6 6 8 5 5 10 8 8 ...
##  $ SosMed_Duration                        : num  3 3 2 5 3 1 4 5 2 5 ...
##  $ SosMed_Medium                          : Factor w/ 13 levels "Linkedin","Youtube",..: 1 2 1 3 3 2 3 3 4 3 ...
##  $ WatchingTV_Duration                    : num  1 0 0 0 1 0 0 0 1 3 ...
##  $ NumberOfMeals                          : num  4 3 3 3 4 1 3 3 3 3 ...
##  $ WeightChange                           : Factor w/ 3 levels "Decreased","Remain Constant",..: 3 1 2 1 2 1 3 3 3 1 ...
##  $ Having_HealthIssue                     : Factor w/ 2 levels "NO","YES": 1 1 1 1 1 2 1 2 1 2 ...
##  $ Stress_Busters                         : Factor w/ 86 levels "Cooking","Scrolling through social media",..: 1 2 3 4 5 6 4 2 7 8 ...
##  $ Time_Utilized                          : Factor w/ 2 levels "NO","YES": 2 2 1 1 1 1 1 1 1 1 ...
##  $ Connected_with_Family_Friends_Relatives: Factor w/ 2 levels "NO","YES": 2 1 2 1 1 2 2 2 1 1 ...
##  $ Most_Missed_Things                     : Factor w/ 36 levels "School/college",..: 1 2 3 4 3 1 4 5 4 1 ...

Now our data is reduced to 1048 observations

# reshape the data into a long format
df_long <- df %>%
  pivot_longer(
    cols = c(OnlineClass_Time, SelfStudy_Duration, Fitness_Duration, Sleeping_Duration, SosMed_Duration, WatchingTV_Duration),
    names_to = "Activity",
    values_to = "Duration"
  )

p_combined <- ggplot(df_long, aes(x = Having_HealthIssue, y = Duration, fill = Having_HealthIssue)) +
  geom_boxplot(color="black") +
  coord_flip() +
  labs(title="Time spent on Various Activities by Health issue during lockdown", 
       x="Health issue during lockdown", 
       y="Time spent") +
  facet_wrap(~ Activity) +
  scale_fill_manual(values = c("YES" = "pink", "NO" = "purple")) +
  theme(legend.position = "none")

print(p_combined)

The study did not establish a straightforward connection between activity levels and health issues, suggesting that multiple factors influence students’ health, such as mental well-being, diet, and pre-existing medical conditions. To comprehensively grasp these complex dynamics, a thorough approach is necessary.

meal_health_plot <- ggplot(df, aes(x = Having_HealthIssue, y = NumberOfMeals)) +
  geom_violin(color="black", fill="gray") +
  coord_flip() +
  labs(title="Number of Meals a Day by Health issue during lockdown", 
       x="Health issue during lockdown", 
       y="Number of Meals a Day") +
  scale_fill_brewer(palette="Set1") + # set color palette to fill the graph
  theme(legend.position = "none")
print(meal_health_plot)

• Majority of the respondents eat 3 times per day, followed by 2 times and 4 times per day
• Respondents with no health issue tend to eat more, than the one who’s health issue

weight_change_plot <- ggplot(df, aes(x=NumberOfMeals, y=WeightChange, fill=WeightChange)) +
  geom_density_ridges() +
  labs(title = "Change in Weight vs. Number of Meals per Day",
       x = "Number of Meals per Day",
       y = "Change in Weight",
       fill = "Change in Weight") +
  theme(axis.text.y = element_blank()) # hide the y-axis because it isn't used in this graph

print(weight_change_plot)

• Respondents who eat less than 3 times a day tend to lose weight Respondents who eat more than 3 times a day tend to gain weight
• Most respondents who eat 3 times a day maintain a stable weight.

Next, we’ll look for anomalies

FindOutliers(df$NumberOfMeals)$summary

##        method    n nMiss nOut    lowLim    upLim minNom maxNom
## 1  ThreeSigma 1048     0    3 0.5099625 5.306831      1      5
## 2      Hampel 1048     0  505 3.0000000 3.000000      3      3
## 3 BoxplotRule 1048     0   45 1.5000000 4.500000      2      4

The first column lists the name of the outlier detection method used. The second column shows the total number of observations, and the third column indicates how many observations are missing (recorded as NA). The fourth column displays the number of outliers detected by each method. The next two columns present the lower and upper limits used to identify outliers. The final two columns show the lower and upper boundaries of the data values that are not considered outliers.

For example, the three-sigma rule identified only 3 outliers, while the Hampel method found 505 outliers, and the boxplot rule identified 45 outliers.

FindOutliers(df$OnlineClass_Time)$summary

##        method    n nMiss nOut    lowLim    upLim minNom maxNom
## 1  ThreeSigma 1048     0    0 -2.879921 9.336696    0.0      9
## 2      Hampel 1048     0   31 -1.447800 7.447800    0.0      7
## 3 BoxplotRule 1048     0  107  0.500000 9.500000    0.5      9

It’s evident that the three-sigma rule detected no outliers, whereas the Hampel method identified 31 outliers, and the boxplot rule flagged 107 outliers.

FindOutliers(df$SelfStudy_Duration)$summary

##        method    n nMiss nOut    lowLim    upLim minNom maxNom
## 1  ThreeSigma 1048     0    0 -2.079668 7.337301      0      7
## 2      Hampel 1048     0   13 -2.447800 6.447800      0      6
## 3 BoxplotRule 1048     0   72  1.000000 7.000000      1      7

It’s clear that according to the analysis: * The three-sigma rule found 0 outliers * The Hampel method identified 13 outliers * The boxplot rule detected 72 outliers

FindOutliers(df$Fitness_Duration)$summary

##        method    n nMiss nOut    lowLim    upLim minNom maxNom
## 1  ThreeSigma 1048     0    0 -1.179393 2.620614      0    2.5
## 2      Hampel 1048     0    0 -1.223900 3.223900      0    2.5
## 3 BoxplotRule 1048     0    0 -0.500000 2.500000      0    2.5

It’s apparent that all the rules identified 0 outliers.

FindOutliers(df$Sleeping_Duration)$summary

##        method    n nMiss nOut   lowLim    upLim minNom maxNom
## 1  ThreeSigma 1048     0    0 3.233145 12.46304      4     12
## 2      Hampel 1048     0    0 3.552200 12.44780      4     12
## 3 BoxplotRule 1048     0   41 6.000000 12.00000      6     12

We can observe that both the three-sigma and Hampel rules identified 0 outliers, while the boxplot rule flagged 41 outliers.

FindOutliers(df$SosMed_Duration)$summary

##        method    n nMiss nOut    lowLim    upLim minNom maxNom
## 1  ThreeSigma 1048     0    0 -1.913046 6.248638      0      6
## 2      Hampel 1048     0    0 -2.447800 6.447800      0      6
## 3 BoxplotRule 1048     0    0  0.000000 6.000000      0      6

It’s apparent that all the rules identified 0 outliers.

FindOutliers(df$WatchingTV_Duration)$summary

##        method    n nMiss nOut    lowLim   upLim minNom maxNom
## 1  ThreeSigma 1048     0   11 -2.204534 4.13603      0      4
## 2      Hampel 1048     0    0 -3.447800 5.44780      0      5
## 3 BoxplotRule 1048     0    0 -1.000000 5.00000      0      5

According to the three-sigma rule, there are 11 outliers detected, while neither the Hampel nor the boxplot rules found any outliers.

Now, we’ll proceed to remove the outliers identified by the three-sigma rule.

# function to remove outliers using three-sigma rule: data outside the range of mean +-3 × standard deviation are considered outliers
remove_outliers_3sigma <- function(df, col_name) {
  mean_val <- mean(df[[col_name]], na.rm = TRUE) # calculate mean with NA value removed
  sd_val <- sd(df[[col_name]], na.rm = TRUE) # calculate std with NA value removed
  
  lower_limit <- mean_val - 3 * sd_val
  upper_limit <- mean_val + 3 * sd_val
  
  # keep the rows where the values in the specified column (col_name) are within the lower and upper limit
  df_filtered <- df[df[[col_name]] >= lower_limit & df[[col_name]] <= upper_limit, ] #buat ngasi syarat hrs ada koma [,]
  
  return(df_filtered)
}

# remove outliers using our function
df <- remove_outliers_3sigma(df, "NumberOfMeals")
df <- remove_outliers_3sigma(df, "NumberOfMeals")
df <- remove_outliers_3sigma(df, "OnlineClass_Time")
df <- remove_outliers_3sigma(df, "SelfStudy_Duration")
df <- remove_outliers_3sigma(df, "Fitness_Duration")
df <- remove_outliers_3sigma(df, "Sleeping_Duration")
df <- remove_outliers_3sigma(df, "SosMed_Duration")
df <- remove_outliers_3sigma(df, "WatchingTV_Duration")

Numerical Variables

1. Shapiro-Wilk Test for Normality

Before conducting the Pearson correlation, we performed the Shapiro-Wilk test to assess the normality of the numerical variables. None of the variables passed the normality test (all p-values < 0.05), indicating that the data does not follow a normal distribution. Therefore, Pearson correlation, which requires normality, cannot be applied in this case.

col_names = c()
sph_test = c()
for (i in num_cols)
{
  col_names <- c(col_names, i)
  ans <- shapiro.test(df[[i]])$p.value
  sph_test <- c(sph_test, ans)
}
sph_df = data.frame(Name = col_names, PVal = sph_test)
sph_df[order(sph_df$PVal),]

##                  Name         PVal
## 4    Fitness_Duration 6.107004e-34
## 1                 Age 3.445997e-32
## 7 WatchingTV_Duration 1.407972e-31
## 8       NumberOfMeals 9.956070e-30
## 6     SosMed_Duration 3.087665e-25
## 5   Sleeping_Duration 6.143239e-21
## 3  SelfStudy_Duration 1.334916e-20
## 2    OnlineClass_Time 3.777746e-16

None of the variables meet the criteria for normal distribution, as all p-values are less than 0.05. Therefore, we can conclude that our numerical data does not follow a normal distribution, which means we cannot use Pearson correlation.

2. Correlation Analysis

num_df <- subset(df,select = num_cols) #update the num_df after removing some of the observations

# convert YES/NO factor columns to numeric (1 for YES, 0 for NO)
num_df$Having_HealthIssue <- as.numeric(df$Having_HealthIssue == "YES")
num_df$Time_Utilized <- as.numeric(df$Time_Utilized == "YES")
num_df$Connected_with_Family_Friends_Relatives <- as.numeric(df$Connected_with_Family_Friends_Relatives == "YES")

# convert WeightChange to numeric based on mappings
num_df$WeightChange <- ifelse(df$WeightChange == "Increased", 2,
                          ifelse(df$WeightChange == "Decreased", 0, 1))

c <- cor(num_df)
# View(c)
corrplot(c, type="upper", method="number",
         number.cex=0.7,  # font size of correlation numbers
         tl.cex=0.7,      # font size of column/row names
         tl.srt=60)

• Calculated the correlation matrix (cor) for numerical variables
• Visualized correlations using corrplot to analyze relationships between variables.

# check correlation of having health issue category with quantitative data which is duration of different activities
for (i in duration_cols)
{
  cat(i,": ",cor(num_df[[i]], num_df$Having_HealthIssue),"\n")
}

## OnlineClass_Time :  -0.113825 
## SelfStudy_Duration :  -0.01142817 
## Fitness_Duration :  -0.04178206 
## Sleeping_Duration :  -0.0176028 
## SosMed_Duration :  0.01514649 
## WatchingTV_Duration :  -0.03181608

There tends to be a mild correlation between the time spent on different activities and the likelihood of experiencing health issues. Longer durations spent on online classes, fitness activities, and watching TV show slight negative associations with health problems. Conversely, spending more time on social media shows a negligible positive correlation with health issues. These correlations, however, are generally weak, suggesting that individual activities may not strongly predict the likelihood of having health issues. It’s likely that other factors not measured in this study also influence these relationships.

# compute the total time for productive and non productive activity
num_df$TotalP = num_df$SelfStudy_Duration + num_df$Fitness_Duration
num_df$TotalN = num_df$SosMed_Duration + num_df$WatchingTV_Duration
# View(num_df)

# compute correlation
correlations <- cor(num_df[, c("TotalP", "TotalN", "Time_Utilized")])
print(correlations)

##                   TotalP      TotalN Time_Utilized
## TotalP         1.0000000 -0.18156524    0.23649549
## TotalN        -0.1815652  1.00000000   -0.09479337
## Time_Utilized  0.2364955 -0.09479337    1.00000000

ggplot(num_df, aes(x = TotalP, 
                     y = Time_Utilized)) +
  geom_point(color="green4") +
  labs(title = "Correlation Between SelfStudy and Fitness Duration with Time_Utilized")+
  geom_smooth(method="lm")

As the combined time spent on productive activities like self-study and fitness increases, there is a slight tendency for the overall time utilized to also increase.

correlations <- cor(num_df[, c("NumberOfMeals", "WeightChange")])
print(correlations)

##               NumberOfMeals WeightChange
## NumberOfMeals     1.0000000    0.1141691
## WeightChange      0.1141691    1.0000000

ggplot(num_df, aes(x = NumberOfMeals, 
                     y = WeightChange)) +
  geom_point(color="green4") +
  labs(title = "Weight Change by Number Of Meals")+
  geom_smooth(method="lm")

This suggests a weak positive relationship: as the number of meals consumed increases, there is a slight tendency for weight change to also increase.

Categorical Variables

1. Chi-Square Test

• Chi-square tests were performed between all pairs of categorical variables to examine their association.
• The p-value for each pair was calculated using the chisq.test function to assess the significance level of 0.05. A p-value less than 0.05 suggests rejection of the null hypothesis (indicating variables are independent).

col_pair = c()
chisq_pv = c()
# harus dikasi ( ) klo ga nanti error
for (i in 2:(length(cat_cols)-1))
{
  for (j in (i+1):length(cat_cols))
  {
    col_pair <- c(col_pair,paste(cat_cols[i],",", cat_cols[j]))
    ans <- chisq.test(df[[cat_cols[i]]], df[[cat_cols[j]]])$p.value
    chisq_pv <- c(chisq_pv, ans)
    # print(chisq.test(df[[cat_cols[i]]], df[[cat_cols[j]]])$p.value)
  }
}

chi_df = data.frame(Name = col_pair, PVal = chisq_pv)
head(chi_df[order(chi_df$PVal),], n=10)

##                                                            Name          PVal
## 33                          Stress_Busters , Most_Missed_Things 3.969857e-274
## 18                               SosMed_Medium , Stress_Busters 6.805568e-233
## 8                       OnlineClass_Rating , Most_Missed_Things  1.042444e-63
## 5                           OnlineClass_Rating , Stress_Busters  1.740276e-33
## 21                           SosMed_Medium , Most_Missed_Things  1.490637e-22
## 15                      OnlineClass_Medium , Most_Missed_Things  5.783002e-21
## 6                            OnlineClass_Rating , Time_Utilized  3.927739e-17
## 34      Time_Utilized , Connected_with_Family_Friends_Relatives  4.941566e-10
## 12                          OnlineClass_Medium , Stress_Busters  2.114716e-07
## 7  OnlineClass_Rating , Connected_with_Family_Friends_Relatives  6.117431e-06

Some variables show a strong connection because we reject the null hypothesis (which states the variables are independent). In statistical terms, we reject the null hypothesis when the p-value is less than 0.05. This indicates that these two variables are not independent in the population from which this sample was taken.

2. Association Statistics

• We calculated association statistics such as the Chi-square test statistic, Cramér’s V, and contingency coefficients for specific pairs of variables.

tbl1 <- table(df$OnlineClass_Rating, df$Time_Utilized) # creates a contingency table: a frequency table that displays the distribution of one categorical variable in rows and another in columns
print(PercTable(tbl1))

##                             
##                     NO   YES
##                             
## Very poor   freq   239   134
##             perc 23.1% 13.0%
##                             
## Poor        freq    14    11
##             perc  1.4%  1.1%
##                             
## Average     freq   174   162
##             perc 16.8% 15.7%
##                             
## Good        freq    89   126
##             perc  8.6% 12.2%
##                             
## Excellent   freq    11    74
##             perc  1.1%  7.2%
## 

print(PercTable(tbl1, margin=1)) # sum of percentage per row

##                                   
##                     NO   YES   Sum
##                                   
## Very poor   freq   239   134   373
##             perc 23.1% 13.0% 36.1%
##                                   
## Poor        freq    14    11    25
##             perc  1.4%  1.1%  2.4%
##                                   
## Average     freq   174   162   336
##             perc 16.8% 15.7% 32.5%
##                                   
## Good        freq    89   126   215
##             perc  8.6% 12.2% 20.8%
##                                   
## Excellent   freq    11    74    85
##             perc  1.1%  7.2%  8.2%
## 

print(PercTable(tbl1, margin=2)) # sum of percentage per column

##                             
##                     NO   YES
##                             
## Very poor   freq   239   134
##             perc 23.1% 13.0%
##                             
## Poor        freq    14    11
##             perc  1.4%  1.1%
##                             
## Average     freq   174   162
##             perc 16.8% 15.7%
##                             
## Good        freq    89   126
##             perc  8.6% 12.2%
##                             
## Excellent   freq    11    74
##             perc  1.1%  7.2%
##                             
## Sum         freq   527   507
##             perc 51.0% 49.0%
## 

Respondents who spent more time utilizing their time generally gave more positive ratings compared to those who spent less time.

# computes a variety of association statistics for the contingency table, such as Chi-square test statistic, Cramér's V, and contingency coefficients, to determine the strength and significance of the association between the two categorical variables
assocstats(tbl1)

##                     X^2 df P(> X^2)
## Likelihood Ratio 89.102  4        0
## Pearson          83.052  4        0
## 
## Phi-Coefficient   : NA 
## Contingency Coeff.: 0.273 
## Cramer's V        : 0.283

tbl1_plt = ggplot(df, aes(x = OnlineClass_Rating, fill = Time_Utilized)) + 
  geom_bar(position = position_dodge(preserve = "single")) + # add a bar plot where bars of different categories are placed side by side, but bars with only one value will remain centered without dodging
  labs(y = "Frequency (in tens)", 
       title = "Online Class Rating by Time Utilized",
       fill = "Time Utilized?") +
  scale_y_continuous(labels = function(x) x / 10)
print(tbl1_plt)

There are no strict rules for interpreting Cramer’s V values, but as a guide: weak (0 to 0.2), moderate (0.2 to 0.3), strong (0.3 to 0.5), and redundant (0.5 to 0.99, indicating the variables likely measure the same concept).

In this case, there is a moderate association between the online class rating and time utilized.

tbl2 <- table(df$Connected_with_Family_Friends_Relatives, df$Time_Utilized)
print(PercTable(tbl2))

##                       
##               NO   YES
##                       
## NO    freq   201   103
##       perc 19.4% 10.0%
##                       
## YES   freq   326   404
##       perc 31.5% 39.1%
## 

print(PercTable(tbl2, margin=1))

##                             
##               NO   YES   Sum
##                             
## NO    freq   201   103   304
##       perc 19.4% 10.0% 29.4%
##                             
## YES   freq   326   404   730
##       perc 31.5% 39.1% 70.6%
## 

print(PercTable(tbl2, margin=2))

##                       
##               NO   YES
##                       
## NO    freq   201   103
##       perc 19.4% 10.0%
##                       
## YES   freq   326   404
##       perc 31.5% 39.1%
##                       
## Sum   freq   527   507
##       perc 51.0% 49.0%
## 

assocstats(tbl2)

##                     X^2 df   P(> X^2)
## Likelihood Ratio 40.127  1 2.3801e-10
## Pearson          39.554  1 3.1906e-10
## 
## Phi-Coefficient   : 0.196 
## Contingency Coeff.: 0.192 
## Cramer's V        : 0.196

tbl2_plt = ggplot(df, aes(x = Connected_with_Family_Friends_Relatives, fill = Time_Utilized)) + 
  geom_bar(position = position_dodge(preserve = "single")) +
  labs(y = "Frequency (in hundreds)", 
       title = "Connected with Family Friends Relatives? by Time Utilized",
       fill = "Time Utilized?") +
  scale_y_continuous(labels = function(x) x / 100)
print(tbl2_plt)

The findings show a connection between effective time management and stronger family bonds. Students who managed their time well reported closer relationships with their families. This is important as it highlights how good time management can enhance personal connections during challenging circumstances.